Intermediate Spatial Analysis in R - Day 2 - section 2
carverd@colostate.edu; contributors @ github (ecarcelen, habpippin, ccmothes)
What Do Nighttime Lights Tell Us About a Place?
Nighttime lights are a uniquely human phenomenon. Every night we capture data on this form of human emissions with the Earth Observation System VIIRS. While we expect that this data will tell us a great deal about the people living and working beneath the lights, there are still many factors we need to evaluate to back up those assumptions.
Question: Could this person be affecting the monthly averages for a dark location? What if the night this image was taken was the only cloud-free capture for that month?
Click for an Answer
It’s unlikely that a headlamp would be seen from space, but we don’t know that for sure. We know that the random variability that events like this represent will be more impactful when fewer observations are present. Rather than try to answer these questions, it might be better to respect that the world is complex. Any automated task (like the one we are working on) will contain conditions that you are probably unaware of and couldn’t do much about, even if you did know they existed. Those assumptions, known and unknown, carry through the analysis and have the potential to skew your results in unexpected ways. Knowing that there is a lot you don’t know helps you keep your mind keen for more questions rather than set on a specific answer.
Where Did We Leave Off?
At the end of the last lesson, we created a correlation between
yearly average radiance and poverty/age at the census tract level for
three different counties in Texas. While the trends were not convincing,
this led us to consider the quality of the input data, which we will
explore in this lesson.
Question: Is what we’ve seen so far still interesting enough to continue work?
Click for an Answer
Hopefully, the answer is yes because this is an educational tutorial, after all. Yet, your answer may feel differently if you were exploring this in a professional context. Looking into one question means you can’t investigate another. It is important to be aware of the Sunk Cost Fallacy and loss aversion as it affects your research. The more we invest in something, the less likely we are to back out.
Counts Data
The average monthly radiance values are delivered with an associated image layer that reports the number of daily observations that were used to generate the mean monthly value for each observation location. These values can range from 0-31, depending on the month. We want to ensure that we are only looking at observation locations where we have a high degree of confidence that the values represent a true mean.
There are three reasons why this is worth evaluating.
View Angle
The VIIRS
sensor captures data in a 3,000 km swath. This means there are a lot
of areas that are off-nadir captures. As night lights are generally
directional features (think lights on the side of the building), the
angle at which the image is captured will affect the radiance observed.
This means that some nightly images will have lower or higher values
than the actual observed value at nadir. We can get around this problem
by working only with on nadir passes, but those images occur only every
14 days or so. That timeframe severely limits our total number of
observations and would require a whole new workflow based on the daily
images. The second option would be to ensure you have enough
observations to average out that variability. How much is enough? We’ll
try to find out.
Lunar Radiance
The moon is a large roundish
rock that has influenced culture and inspired people’s curiosity for as
long as we’ve been around. We’ve always cared about the moon because it
reflects the radiance of the sun on the Earth. This reflectance means
that we can see it on an otherwise dark night. Due to the orbit patterns
of these three celestial bodies, we end up with about half of each month
being darker than the other half. The effects of lunar radiance are
substantial in darker areas, especially when combined with freshly
fallen snow. Our counts data does not tell us anything
about the date of the image captured, but we’ll cross our fingers and
assume it’s normally distributed. Therefore, the more observations we
get, the less we need to worry about the lunar radiance making places
look brighter than they are.
